esa-presentation

Raising the Bar:

Learning and Teaching Better

Data Visualization in R

Presented by Marney Pratt & June Arriens

Contributions by Audrey McCombs & Dan Turner

August 16, 2022

Raise The Bar Interactive Website: https://raisethebar.netlify.app/

email questions to: mcpratt@smith.edu

Why Bar Graphs Are Misleading

Go to Chapter 1 then click on (1) Introduction

https://raisethebar.netlify.app/chapter1

How does the population of Elliptio complanata mussels differ through time and location?

This image shows a map of where mussels were sampled and what an Elliptio complanata mussel looks like

Smith College faculty and students sampled Elliptio complanata mussels in two rivers in Western Massachusetts in June of 2016-2019 to determine whether or not the college’s management of sediment in Paradise Pond was impacting the mussels. At each sampling time, we set up a 10m by 30m area in the river and then chose x and y coordinates of predetermined number of random starts (in this case 3) using a systematic sampling technique (pages 10-18 in Strayer and Smith 2003). We placed a 50 cm x 50 cm quadrat at each set of coordinates, looked with a view bucket to see if any mussels were at the surface, then we excavated the substratum with a trowel or gloved hands down to about 15 cm. The excavated substrate was passed through a sieve with 4.67 mm mesh. All mussels found were identified to species, counted (also noted if on the surface or under the surface, alive or dead), and had their length measured to the 0.1mm.

Mussel Density

Table showing the locations (Manhan and Mill), years (2016-2019), number of quadrats sampled (21-56 depending on the year and location), and density of Elliptio complanata mussels in number per meters squared

bar graph of mean shell lengths with standard errors — Let’s answer some questions on the website about mussel size using this bar graph:

https://raisethebar.netlify.app/chapter1

Bar Graphs with Means & Error Bars Can Be Misleading

Does not show the distribution, range, or sample size of the actual values
Misleads because of common belief that values fall within the bar rather than above
You can get the same bars with very different distributions

See Weissgerber et al. 2019 paper from the journal Circulation

Bar graphs do not allow us to answer questions about

distribution,
sample size, or
range

Are there other graph types that can help use answer these questions more effectively?

Resources for Other Useful Graph Types

Note

We use the term “dot plot” here, but a plot that shows all the points for different groups can also be called a beewarm-stye plot, jitter plot, violin scatter plot, column scatter plot, jittered strip plot, jittered individual value plot, among others.

Mussel Length Data Graphed in 4 Ways

This image shows 4 different plot types with the same mussel length data. Bars that show means and standard errors, jittered dot plots with medians that show the distribution of the data, box plots, and violin plots.

Here we see the same Elliptio complanata mussel length data graphed in four different ways:

Bar plots showing the means and standard errors,
Dot plots showing individual points jittered to show the data distribution and the median shown as the horizontal bar,
Box plots showing the median (middle horizontal bar), middle 50% of the data (= the interquartile range shown as the box), whiskers (reach up to 1.5 times the interquartile range from the edges of the box), and outliers (points shown outside of the whiskers), and
Violin plots which show the distribution of the data as a mirrored, smoothed probability density curve. Go back to the previous slide and click on the resource links if you need help interpreting what these plots show. What are the pluses and minuses of each of these graph types?

Visualizing Grouped Continuous Data

Show all points or distribution when possible
Distribution of data & sample size determine summary statistics to use
- Small samples size makes summary stats less reliable
- Only use mean & SD if normal
- Box plots should NOT be used for multimodal distribution

What Graph to Choose When You Have a

Categorical Independent Variable

This image shows a table to help choose a good graph type. See the slide notes on the website for a thorough description. — (adapted from Weissgerber et al. 2019 by Marney Pratt, March 3, 2022)

Mix and Match When Needed

This image shows a graph of the mussel lengths that mixes a violin, box, and dot plot.

It can be helpful to mix and match different graph types to get the benefits of each to compliment each other. Here we have mixed a dot plot, box plot, and violin plot. The sample size for the number of mussels from each year in each river has also been added.

In this particular case, the distribution of mussel shell lengths is multimodal (several clumps), which suggests different size/age classes of mussels.

Since we moved sediment upstream of the mussels in the Mill River later in the summer of 2016, the 2016 samples collected early in the summer are considered our “before” samples, while the other years are “after” sediment redistribution. To see a negative impact of sediment redistribution on the mussels in the Mill River, we might expect to see a decrease in mussel density or size in the Mill River (our “impact” site) compared to the Manhan River (our “control” site) after sediment is moved.

It looks like there was a better recruitment of small mussels in 2018 in the Mill River. Note the median looks slightly higher in the Manhan River than the Mill River in 2019, but there are a lot more small mussels as well as some larger mussels in the Mill River. Since we find the mussel density has increased (see Table 1 in the chapter 1 Introduction slide set) and there is a good mix of sizes of mussels (including those that are smaller and larger than in the Manhan River). We don’t see any clear evidence of a negative impact of managing sediment upstream in Paradise Pond on the Mill River mussels.

Next we will practice using template code to make these different graphs using R

Why use R?

It’s free, open source
TONS of resources & flexibility
- Lots of packages & functions specific to ecology
- Online R community
- Cool graphs: R Graph Gallery
Reproducible
- Many journals now require code for analysis to be shared
- Everything (clean, analyze, plot, communicate data) all in one
Highly transferable skill
Used by an increasing number of ecologists

Can You Learn R or Teach it to Students with NO Experience?

YES!

Start with code templates to make different graph types

Tip

Manipulating data (“data wrangling”) is much harder than graphing. For people just starting in R, we recommend doing the data wrangling in a spreadsheet program or use already prepared data. Beginners will have a better experience just using R for graphing with template code.

Let’s Make some Graphs!

https://raisethebar.netlify.app/chapter2

Tip

During very first exposures, it can be helpful if the learner doesn’t have to install anything - not R, RStudio, or packages.

The “Raise the Bar” website is one option for exposing novices to how to make some different graphs using R code templates without having to install anything. All you need is a browser and internet access!

When learning as a novice or teaching novices, it can be helpful to start with most of the code for a particular graph type already written with just some blanks for the data, variables, and labels. Then users can fill in the blanks and tweak the code as desired if they want to try changing other aspects (colors, size, fonts, etc).

During very first exposures, it can be helpful if the learner doesn’t have to install anything - not R or RStudio (the two software packages needed to run R locally on your computer) and not any packages (packages that can expand the capabilities of base R to do all kinds of things are often needed and must be installed before first use).

The “Raise the Bar” website is one option for exposing novices to how to make some different graphs using R code templates without having to install anything. All you need is a browser and internet access.